What are some popular consumer products?

What are the best practices for forecasting demand for a startup in tech or consumer products?  What is your strategy for logically building the forecast?

• This is applicable both to technology startups (forecasting demand as the cost driver for variable costs like EC2) and consumer products like fashion or apparel. Now, this question probably lends itself to getting into a complex discussion about Bass, Diffusion, and Bayesian applications.  I am looking more for the logic for determining a defensible demand forecast. For a Technology SaaS Application: How do you forecast site traffic growth when you are limited to imperfect information from Quantcast & Compete (whose results are often very different)? For a consumer product company: Traditionally defining the # of retailers in each given territory and then multiplying by a projected weighted average sku. What are some of the practices & resources that you have found extremely useful for forecasting demand in a start without historical information?

• Answer:

  I have never worried about calculating site traffic to forecast variable technology costs because--unless you're not going to be making money from the traffic--the variable technology costs are going to be so small compared to revenue.  (This also depends upon designing your cloud infrastructure properly, but if you have the right people, you'll design it properly). In my experience in launching several large cloud infrastructures (the largest being upwards of $250K/year in infrastructure costs), the "cost" pain has much more to do with rapid development and testing early on that once you're delivering your product to the world. For example, BuildFax's product delivery servers cost roughly $2400/month to host on EC2, and we can deliver more than 10 million reports per month on that infrastructure.  The average price we charge for a report is around $2.  However, it cost us well into six figures (of EC2/EBS charges) to do all of the development and testing to get the servers to that level. There is no reason with proper architecture (optimized application and database servers + memcached + s3 + cloudfront) that any application but the largest (e.g. Google, Facebook) will need to worry intensely about infrastructure costs if money is coming in.  I would draw the "largest" line at around 5 million unique visitors per month.  Anything below that and if you do it right, you can handle the traffic on a small number of servers.